bohlt/dagzoo-default-baseline-demo-20260407

---
pretty_name: Default Baseline Synthetic Tabular Corpus
tags:
- tabular
- synthetic
- dagzoo
license: apache-2.0
---

# Default Baseline Synthetic Tabular Corpus

This dataset repository contains synthetic tabular corpora generated with `dagzoo`.
Balanced mixed-type baseline with the default factorized complete-data feature prior plus latent-complete-data target-head semantics.

## What is included

- `generated/`: public parquet shards plus per-shard `dataset_catalog.ndjson` summaries
- `curated/`: optional accepted-only shards written later by `dagzoo filter`
- `handoff_manifest.json`: portable corpus identity and provenance metadata

## Corpus summary

- generated datasets: 2
- task mix: classification (2)
- train rows per dataset: 768
- test rows per dataset: 256
- features per dataset: 16 to 49
- classes per classification dataset: 4 to 5
- observed feature types: cat, num

## Generation provenance

- source family: `dagzoo.heterogeneous_scm`
- generate run id: `98276f918489ba560f149fa887820470`
- generated corpus id: `e8bc35defe4260a3118f3cee1ee08853`
- config reference: `recipe:default-baseline`
- target derivation: `tabiclv2_latent_node`
- missingness: `none`

## Publishing workflow

This repo is produced from a local handoff root with:

```bash
dagzoo generate --config recipe:<name> --num-datasets <n> --handoff-root handoffs/<run_name>
dagzoo publish hub --handoff-root handoffs/<run_name> --repo-id <namespace/name>
```

Only the public handoff artifacts are uploaded. Local `internal/` sidecars stay on disk for dagzoo tooling and are not published to the Hub.

## Links

- dagzoo repo: [github.com/bensonlee5/dagzoo](https://github.com/bensonlee5/dagzoo)
- dagzoo docs: [bensonlee5.github.io/dagzoo/docs/](https://bensonlee5.github.io/dagzoo/docs/)

## References

- Dagzoo packaged baseline recipe.